Some thoughts... David Rush (21 Nov 2001 19:26 UTC)
Bad things Re: Some thoughts... Jussi Piitulainen (21 Nov 2001 20:25 UTC)
Re: Bad things Re: Some thoughts... David Rush (22 Nov 2001 16:10 UTC)
Access time of elements Re: Bad things [] Jussi Piitulainen (27 Nov 2001 10:59 UTC)
Re: Access time of elements Re: Bad things [] Per Bothner (27 Nov 2001 17:10 UTC)
Re: Access time of elements Re: Bad things [] David Rush (27 Nov 2001 17:25 UTC)
Re: Access time of elements Re: Bad things [] Per Bothner (27 Nov 2001 17:55 UTC)
Re: Access time of elements Re: Bad things [] David Rush (27 Nov 2001 21:19 UTC)
Re: Access time of elements Re: Bad things [] Jussi Piitulainen (28 Nov 2001 15:40 UTC)
Re: Access time of elements Re: Bad things [] Jussi Piitulainen (28 Nov 2001 16:20 UTC)
Re: Access time of elements Re: Bad things [] Noel Welsh (28 Nov 2001 10:55 UTC)
Re: Access time of elements Re: Bad things [] Jussi Piitulainen (28 Nov 2001 17:21 UTC)

Re: Access time of elements Re: Bad things [] Noel Welsh 28 Nov 2001 10:55 UTC

--- Per Bothner <xxxxxx@bothner.com> wrote:
> Jussi Piitulainen wrote:
>
> >Shall we switch, and is a vector good enough for an
> index object? (I
> >believe somebody asked for such packaging even
> apart from efficiency.)
> >
> A modest but not order-or-magnitude difference on a
> non-optimizing
> implementation is not does not IMO justify a less
> natural api.

As the "somebody" in question I should explain why I
feel this is a more natural api.  Consider list
processing in Scheme.  Scheme is a 'closed world' for
list processing - you never need operate on anything
but lists because of fold and apply.  I've written
Scheme programs that have no other data structure, and
I'm sure others have.

Now I have two hypotheses:
1)  We want a truly great language for array
processing
2)  We want that language to be Scheme

If that is the case then there needs to be an
array/function duality of the same kind we have for
lists.  I.e. you need to be able to pull apart arrays
and use those parts to pull apart other arrays.  So I
advocate using arrays as indices to arrays (NB: there
is no separate index object in this proposal).

A simple example of why this is desireable:  Imagine
you have a 2-D array storing temparature measured at
space and time, say samples from a number of weather
stations over a period of days.

  temparature := array[location time]

Now imagine that you want to be able to quickly find
out the hotest measurement for each day (maybe you're
working for the fire department and this is important
information).  So you have another array which indexes
into the temparature array.

  hotest := array[time indexes]

This is a 1 x 2 array, where column 0 corresponds to
location and column 1 to time

Now to find the hottest measurement for day 0 you
would do something like this:

  The current proposal:

  (let ((location (array-ref hotest 0 0))
        (time (array-ref hotest 0 1)))
           (array-ref temparatures location time))

  My proposal:

  (array-ref temparatures (array-ref hotest #(0)))

The first array ref returns the array-slice hotest[0:]
The second uses that array slice to index into the
temparatures array.  In my opinion this is much
cleaner code.

I hope that clears things up.

Noel

__________________________________________________
Do You Yahoo!?
Yahoo! GeoCities - quick and easy web site hosting, just $8.95/month.
http://geocities.yahoo.com/ps/info1